Research map / Label-efficient and robust learning / Semi-supervised learning
TransMatch: A Transfer-Learning Scheme for Semi-Supervised Few-Shot Learning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use TransMatch with 100 unlabeled images per class instead of Activation Net | miniImageNet 5-way 1-shot classification · miniImageNet | accuracy (%): 59.60$\pm$0.41 -> 63.02$\pm$1.07 | Helped |
| use TransMatch with 200 unlabeled images per class instead of TADAM | miniImageNet 5-way 5-shot classification · miniImageNet | accuracy (%): 76.70$\pm$0.30 -> 82.24$\pm$0.59 | Helped |
| add MixMatch rather than use Pseudo-Label (ablation) | TransMatch framework · miniImageNet 5-way classification, 100 unlabeled images | accuracy (%): 57.01 $\pm$ 1.13 -> 63.02 $\pm$ 1.07; +6.01 | Helped |
| add MixMatch rather than use Pseudo-Label (ablation) | TransMatch framework · miniImageNet 5-way 2-shot classification, 100 unlabeled images | accuracy (%): 70.07 $\pm$ 0.96 -> 72.36 $\pm$ 0.88; +2.29 | Helped |
| add MixMatch rather than use Pseudo-Label (ablation) | TransMatch framework · miniImageNet 5-way 5-shot classification, 100 unlabeled images | accuracy (%): 80.00 $\pm$ 0.66 -> 81.19 $\pm$ 0.59; +1.19 | Helped |
| add weight imprinting to MixMatch (ablation) | TransMatch on miniImageNet · miniImageNet 5-way 1-shot classification, 100 unlabeled images | accuracy (%): +11.02 | Helped |
| add MixMatch to Imprinting (ablation) | TransMatch on miniImageNet · miniImageNet 5-way 1-shot classification | accuracy (%): 58.68 $\pm$ 0.81 -> 63.02$\pm$1.07 | Helped |
| fine-tune the imprinted classifier without unlabeled examples (ablation) | miniImageNet 5-way 1-shot classification · miniImageNet | accuracy (%): 58.68$\pm$0.81 -> 55.60$\pm$0.77 | Hurt |
| increase unlabeled data from 20 to 100 images (ablation) | TransMatch · miniImageNet 5-way 1-shot classification | accuracy (%): 58.43$\pm$0.93 -> 63.02$\pm$1.07 | Helped |
| increase unlabeled data from 100 to 200 images (ablation) | TransMatch · miniImageNet 5-way 1-shot classification | accuracy (%): 63.02$\pm$1.07 -> 62.93$\pm$1.11 | No clear effect |
| include distractor-class unlabeled data rather than unlabeled data from novel classes (ablation) | MixMatch · miniImageNet 5-way 1-shot classification, 1–3 distractor classes | accuracy (%): 50.14 $\pm$ 1.06; 50.68 $\pm$ 1.15; 49.48 $\pm$ 1.16 | Hurt |
| increase unlabeled data from 5 to 20 images (ablation) | TransMatch · CUB-200-2011 100-way 5-shot classification | accuracy (%): 52.90 -> 59.25 | Helped |
Closest papers by meaning
- A Unified Few-Shot Classification Benchmark to Compare Transfer and Meta Learning Approaches (NeurIPS 2021 · similarity 0.89)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.88)
- Making Large Vision Language Models to Be Good Few-Shot Learners (AAAI 2025 · similarity 0.88)
- OTMatch: Improving Semi-Supervised Learning with Optimal Transport (ICML 2024 · similarity 0.87)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (CVPR 2025 · similarity 0.87)
- Simple Semantic-Aided Few-Shot Learning (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.